1. Field of the Invention
The present invention relates generally to methods and apparatuses for automatic activation of a clock master on a Class II Fast Ethernet repeater stack. In order to enable the repeater stack to comply with the Class II latency requirements in the IEEE 802.3 standard, a synchronous stack bus is provided for the repeaters in the repeater stack. The present invention relates to methods and apparatuses for automatically selecting a repeater from among the group of repeaters in the Fast Ethernet repeater stack to be the repeater that provides a master clock signal for the synchronous bus.
2. Description of the Related Art
The growth of local-area networks (LANs) has been driven by the introduction of Ethernet Technology as well as the availability of powerful, affordable personal computers and workstations. As a result, applications that once were possible only on mainframe computers are now running on LANs. Network speed and availability are critical requirements. However, existing applications and a new generation of multimedia, groupware, imaging, and database products can tax a network running at Ethernet's traditional speed of 10 megabits per second (Mbps). Moreover, with more applications requiring faster LAN speeds for acceptable performance, network managers increasingly find that high-performance computation platforms and mission-critical applications can overwhelm a 10 Mbps network. Network managers therefore are increasingly implementing high-speed LAN technology.
Fast Ethernet
For organizations with existing Ethernet installations, increasing the network speed to 100 Mbps is preferable to investing in a completely new LAN technology. This user preference has driven the industry's decision to specify a higher-speed Ethernet that operates at 100 Mbps. This higher-speed Ethernet is known as Fast Ethernet.
In July 1993, a group of networking companies joined to form the Fast Ethernet Alliance. The charter of the group was to draft the 802.3u 100BaseT specification ("802.3 specification") of the Institute of Electrical and Electronics Engineers (IEEE) and to accelerate market acceptance of Fast Ethernet technology. The final IEEE 802.3 specification was approved in June 1995. Among the other goals of the Fast Ethernet Alliance are: to maintain the Ethernet transmission protocol Carrier Sense Multiple Access Collision Detection (CSMA/CD); to support popular cabling schemes; and to ensure that Fast Ethernet technology will not require changes to the upper-layer protocols and software that run on LAN workstations. For example, no changes are necessary to Simple Network Management Protocol (SNMP) management software or Management Information Bases (MIBs) in order to implement Fast Ethernet.
Other high-speed technologies, such as 100VG-AnyLAN and Asynchronous Transfer Mode (ATM), achieve data rates in excess of 100 Mbps by implementing different protocols that require translation when data moves to and from 10BaseT. Protocol translation requires changing the frame, which often incurs higher latencies when passing through layer 2 (data-link layer) LAN switches.
In many cases, organizations can upgrade to 100BaseT technology without replacing existing wiring. Options for 100BaseT media are the same as those for 10BaseT. They include shielded and unshielded twisted pair (STP and UTP) and fiber. The Media Independent Interface (MII) provides a single interface that can support external transceivers for any of the 100BaseT physical sublayers.
CSMA/CD
Carrier sense-collision detection is widely used in LANs. Many vendors use this technique with Ethernet and the IEEE 802.3 specification. A carrier sense LAN considers all stations as peers; the stations contend for the use of the channel on an equal basis. Before transmitting, the stations monitor the channel to determine if the channel is active (that is, if another station is sending data on the channel). If the channel is idle, any station with data to transmit can send its traffic onto the channel. If the channel is occupied, the stations must defer to the station using the channel.
FIG. 1 depicts a carrier sense-collision detection LAN. Network devices 102, 104, 106, and 108 are attached to a network bus 110. Only one network device at a time is allowed to broadcast over the bus, since if more than one device were to broadcast at the same time, the combination of signals on the bus would likely not be intelligible. For example, assume network devices 102 and 104 want to transmit traffic. Network device 108, however, is currently using the channel, so network devices 102 and 104 must "listen" and defer to the signal from network device 108, which is occupying the bus. When the bus goes idle, network devices 102 and 104 can then attempt to acquire the bus to broadcast their messages.
Because network device 102's transmission requires time to propagate to other network devices, these other network devices might be unaware that network device 102's signal is on the channel. In this situation, network device 102 or 104 could transmit its traffic even if network device 108 had already seized the channel after detecting that the channel was idle. This problem is called the collision window. The collision window is a factor of the propagation delay of the signal and the distance between two competing stations. Propagation delay is the delay that occurs before a a network device can detect that another network device is transmitting.
Each network device is capable of transmitting and listening to the channel simultaneously. When two network device signals collide, they create voltage irregularities on the channel, which are sensed by the colliding network devices. The network devices then turn off their transmission and, through an individually randomized wait period, attempt to seize the channel again. Randomized waiting decreases the chances of another collision because it is unlikely that the competing network devices generate the same wait time.
It is important that the total propagation delay not exceed the amount of time that is required to send the smallest size data frame. This allows devices to discard data corrupted by collisions by simply discarding all partial frames. It is therefore not desirable for entire frames of data to be sent before a collision is detected. Carrier sense networks are usually implemented on short-distance LANs because the collision window lengthens as the channel gets longer. Longer channels provide opportunity for the more collisions and can reduce through-put in the network. Generally, a long propagation delay coupled with short frames and high data transfer rates give rise to a greater incidence of collisions. Longer frames can mitigate the effect of long delay, but they reduce the opportunity for competing stations to acquire the channel.
The IEEE 802.3 specification sets a standard minimum frame size of 64 bytes (512 bits). Therefore, in order for a network to comply with the standard, a station on the network must not be able to transmit 64 bytes of data before a collision is detected.
Although Fast Ethernet maintains CSMA/CD, the Ethernet transmission protocol, it reduces the transmission time for each bit by a factor of 10. Thus, the Fast Ethernet packet speed increases tenfold, from 10 Mbps to 100 Mbps. Data can move between Ethernet and Fast Ethernet without requiring protocol translation or software changes, because Fast Ethernet maintains the 10 BaseT error control functions as well as the frame format and length.
Repeaters
While some Ethernet applications connect numerous network devices to a network bus that is literally a cable connecting the network devices, it is often more desirable to connect network devices using a repeater or hub. It should be noted that in the following description the term "hub" and the term "repeater" are used interchangeably. The repeater manages collision detection for the network devices so that the network devices need only broadcast messages without detecting collisions. The repeater notifies a network device when a collision occurs during its attempt to transmit. In addition, the repeater implements a star topology so that more devices can be included on the network without violating any cable length restriction and so that many devices can be added or removed from the network efficiently.
An Ethernet repeater is a device that serves as a central station for plugging network devices included in an Ethernet network, hence the term "hub." The Ethernet repeater receives messages from the network devices that are plugged into it and broadcasts (or "repeats") the message to all of the other devices on the network along a network bus if no collision is detected. The repeater monitors network traffic in its collision domain and assumes the responsibility for collision detection. The network devices thus simply broadcast messages to the repeater and do not need to first listen before sending messages. If the repeater has already assigned the network bus to a device, then it notifies the device that tried to broadcast that a collision has occurred so that the network device may try again later. The amount of time that it takes for the repeater to receive a data signal and repeat that data signal out to every port on which the data signal is to be broadcast is referred to as the latency of the repeater.
The 802.3 specification contains maximum latency requirements that cannot be exceeded by a conforming repeater. The maximum permissible latency, combined with the requirements for maximum cable length and restrictions on the number and type of other devices allowed within a collision domain, limits the amount of time that it takes to notify a network device that a collision has occurred, ensuring that the overall 802.3 design criteria is met that all collisions are detected before a complete 64 byte frame is transmitted. If the maximum permissible latency were exceeded by a repeater, then multiple devices in the repeater's collision domain on an 802.3 ethernet network might broadcast complete frames of data before being notified of a collision. As described above, the broadcast of complete frames when a collision occurs would defeat a scheme for discarding data associated with collisions by simply discarding all partial frames. Thus, minimizing the latency of a repeater is critical if the repeater is to be implemented on a network in accordance with the 802.3 specification. The 100BaseT standard defines two classes of repeaters: Class I and Class II. At most, a collision domain can include one Class I or two Class II repeaters. Including more than one repeater in a single collision domain is sometimes referred to as cascading repeaters. Specifically, in order to conform to the Class II requirement, the latency a repeater must be less than 46 bit times. It should be noted that the standard is expressed in terms of bit times, or the amount of data that could be transmitted on the network during the latency period.
Network Flexibility
The Class II requirement, which allows more than one repeater to be included in a single collision domain, significantly adds flexibility to network topology. Expanding the number of ports available on a network may be accomplished by simply adding a second repeater in the same collision domain as a single existing repeater. No switch is required. By limiting the latency of the two repeaters, it is ensured that collisions can be detected and devices connected to different repeaters can be notified of collisions in time to stop sending data before a complete frame is broadcast.
Because networks tend to constantly change and expand with network devices being added, it would be highly advantageous if, in addition to the Class II feature of allowing two repeaters in a collision domain, it were also possible that each of the two Class II repeaters were expandable or stackable. Additional ports could be added to a first repeater stack that functions as one Class II repeater and then a second stack could be included as a second Class II repeater. Thus, stackability combined with cascadability would provide even greater flexibility for network expansion.
There is therefore a need for a stackable repeater that could be plugged into a network in a flexible manner along with a group of other repeaters in a stack. For a stack of such repeaters to conform to the latency requirements set forth in the standard, a very high speed, efficient repeater stack bus would be required. Specifically, in order to conform to the Class II requirement, the total latency of all of the repeaters in the stack connected to the repeater stack bus must be less than 46 bit times. A repeater stack synchronous bus for providing a fast connection among a group of stacked repeaters is disclosed in U.S. patent application Ser. No. 08/965,479 which was previously incorporated by reference for all purposes. Because the bus is synchronous, all of the repeaters on the bus are driven by a single master clock signal.
As noted above, one of the primary reasons for providing a repeater stack is to provide flexibility in configuring a network. Repeaters may be added or removed from the stack or repeaters in the stack may be powered on or off as the network is reconfigured. One way to provide a master clock to the repeater stack would be to provide an external clock to the stack. However, it would be more preferable if an internal clock in one of the repeaters could be selected and activated as the clock master for the repeater stack bus. What is needed is a reliable way of selecting a repeater to provide the master clock signal and activating that master clock signal for the repeater stack synchronous bus. Preferably, the master clock signal should be activated and propagated on the bus even when one or more repeaters on the bus are powered off.